Assembling assays into a master MultiAssayExperiment object

library(tidyverse)
library(magrittr)
library(gt)
library(patchwork)
library(SummarizedExperiment)
library(tidySummarizedExperiment)
library(MultiAssayExperiment)

tmp <- fs::dir_map("R scripts/", source)

theme_set(theme_light())

The purpose of this document is to create a MultiAssayExperiment object containing all assays and clinical data generated in the context of the VIBRANT trial at the participant x visit resolution (which means that duplicates were aggregated in previous steps such that for a given assay/variable, there is a single value for each participant and visit combination).

The document is organised as follow:

• First, we load and minimally transform the “clinical” data (i.e., data collected through CRFs).

• Next, we load and minimally transform the SummarizedExperiment objects for each assay (metagenomics, qPCR, 16S amplicon sequencing, Luminex, and flow cytometry).

• Finally, we create the MultiAssayExperiment object by combining the clinical data and assay data. This entails several steps, including creating a mae_colData table that combines the clinical data and assay sample information, ensuring that the uids (unique identifiers) are consistent across all data tables.

At the end, we document which assays are available (= have data) for which participants and visits.

Clinical data (CRF data)

CRF data (raw and QCed)

crf_files <- 
  get_01_output_dir() |> 
  fs::dir_ls() |> 
  str_subset("/01_")

crf_clean_file <- crf_files |> str_subset("01_crf_clean") |> sort(decreasing = TRUE) |> magrittr::extract(1)
crf_raw_file <- crf_files |> str_subset("01_crf_raw") |> sort(decreasing = TRUE) |> magrittr::extract(1)
crf_dict_file <- crf_files |> str_subset("01_crf_plates_dictionary") |> sort(decreasing = TRUE) |> magrittr::extract(1)

load(crf_clean_file, verbose = TRUE)

Loading objects:
  crf_clean

load(crf_raw_file, verbose = TRUE)

Loading objects:
  crf_raw

load(crf_dict_file, verbose = TRUE)

Loading objects:
  crf_plates_dictionary

rm(crf_clean_file, crf_raw_file, crf_dict_file)

Participants table

load(
  crf_files |> str_subset("01_participants_2025") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  participants

load(
  crf_files |> str_subset("01_participants_variable_dictionary") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  participants_variable_dictionary

load(
  crf_files |> str_subset("01_participant_crfs_merged") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  participant_crfs_merged

Visits table

load(
  crf_files |> str_subset("01_visits_2025") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  visits

load(
  crf_files |> str_subset("01_visits_long_2025") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  visits_long

load(
  crf_files |> str_subset("01_visits_crfs_merged") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  visits_crfs_merged

visits |> 
  ggplot() +
  aes(x = study_day, y = pid |> factor(), color = visit_code |> factor()) +
  geom_point() 

We define the uid (unique identifier):

visits <- 
  visits |> 
  mutate(
    uid = str_c(pid, visit_code, sep = "_")
  ) |> 
  select(uid, everything())

Exposures table

load(
  crf_files |> str_subset("01_exposures_") |> sort(decreasing = TRUE) |> magrittr::extract(1),
  verbose = TRUE
  )

Loading objects:
  exposures

Assays: loading the SummarizedExperiment objects

data_dir <- get_01_output_dir()
se_files <- fs::dir_ls(data_dir, regexp = "se_.*\\.rds$") |> sort(decreasing = TRUE)

Metagenomics data

mg_file <- se_files[grepl("mg", se_files)][1]
se_mg <- readRDS(mg_file)
se_mg <- check_se(se_mg)
se_mg

class: SummarizedExperiment 
dim: 779 996 
metadata(3): name date dictionary
assays(5): count count_corr rel_ab_all rel_ab_bact rel_ab
rownames(779): C0022A1 C0059E1 ... Mycoplasma_arginini
  Streptococcus_virus
rowData names(16): taxon_id taxon_label ... strain_origin biose_id
colnames(996): 068100004_0000 068100004_1000 ... EQ05822515 MC1
colData names(35): uid mg_uid ... notes reextraction_status

qPCR data

qPCR_file <- se_files[grepl("03_se_pcr_agg", se_files)][1]
se_qPRC <- readRDS(qPCR_file)
se_qPRC <- check_se(se_qPRC)
se_qPRC

class: SummarizedExperiment 
dim: 16 1029 
metadata(3): description date assay_and_coldata_dictionary
assays(8): n_non_na dilution ... copies_per_swab_LL3_med
  copies_per_swab_LM_med
rownames(16): C0175A1 C0022A1 ... FF00072 16S
rowData names(9): target fluor ... biose_id pcr_plate_ids
colnames(1029): 068100004_0000 068100004_1000 ... EQ05822515 EQ05822516
colData names(7): uid qpcr_sample_type ... vibr_sample_id
  ext_lib_plate_nb

qPCR data for daily swabs (generated post-unblinding)

daily_qPCR_file <- se_files[grepl("30_se_pcr_agg", se_files)][1]
se_daily_qPRC <- readRDS(daily_qPCR_file)
se_daily_qPRC <- check_se(se_daily_qPRC)
se_daily_qPRC

class: SummarizedExperiment 
dim: 16 2990 
metadata(3): description date assay_and_coldata_dictionary
assays(5): n_non_na dilution copies_per_swab_med copies_per_swab_mean
  copies_per_swab_cv
rownames(16): C0175A1 C0022A1 ... FF00072 16S
rowData names(8): target fluor ... strain_origin biose_id
colnames(2990): 068100004_1001 068100004_1002 ... 068200465_1406
  068200465_1407
colData names(9): uid qpcr_sample_type ... pcr_plate_nb
  ext_lib_plate_nb

16S rRNA amplicon sequencing

amplicon_file <- se_files[grepl("16S_agg", se_files)][1]
se_16S <- readRDS(amplicon_file)
se_16S <- check_se(se_16S)
se_16S

class: SummarizedExperiment 
dim: 817 4119 
metadata(4): name date dictionary asv_tax_table
assays(2): counts rel_ab
rownames(817): Lactobacillus_iners Lactobacillus_crispatus ...
  Oliverpabstia_sp000432335 Corynebacterium_ulcerans
rowData names(12): taxon_id taxon_label ... asv_ids
  last_available_taxonomic_level
colnames(4119): 068100004_0000 068100004_1000 ...
  Nuclease_free_water_plate1_Pool10 Unusedswab_C2_plate1_Pool10
colData names(32): uid exclude_sample ... ext_lib_plate_nb resequenced

Luminex data

luminex_file <- se_files[grepl("luminex_agg", se_files)][1]
se_luminex <- readRDS(luminex_file)
se_luminex <- check_se(se_luminex)
se_luminex

class: SummarizedExperiment 
dim: 48 850 
metadata(7): n_plates names ... extremes se_creation_date
assays(7): obs_conc obs_conc_imp ... obs_conc_log10 obs_conc_imp_log10
rownames(48): CTACK Eotaxin ... VEGF b-NGF
rowData names(3): prop_OOR_per_analyte exclude_analyte
  exclude_analyte_reason
colnames(850): 068100004_0000 068100004_1000 ... 068200465_1900
  068200465_2120
colData names(25): uid sample_id ... visit_code_assay location_assay

Flow cytometry data

flow_file <- se_files[grepl("flow", se_files)][1]
se_flow <- readRDS(flow_file)
se_flow <- check_se(se_flow)
se_flow

class: SummarizedExperiment 
dim: 11 356 
metadata(1): gate_mapping
assays(2): count percentage
rownames(11): Live Granulocytes ... DblePosCD4 DblePosCD8
rowData names(3): cell_type parent_cell_type cell_type_label
colnames(356): 068100004_1000 068100004_1100 ... 068200465_1500
  068200465_2120
colData names(5): uid sample machine sample_type control_type

Creating the MultiAssayExperiment object

Assay list

assay_list <- 
  bind_rows(
    tibble(
      mae_assay_name = "mg", 
      se_object_name = "se_mg", 
      assay_label = "Metagenomic"
      ),
    tibble(
      mae_assay_name = "qPCR", 
      se_object_name = "se_qPRC", 
      assay_label = "qPCR weekly swabs"
      ),
    tibble(
      mae_assay_name = "qPCR_daily", 
      se_object_name = "se_daily_qPRC", 
      assay_label = "qPCR home swabs"
      ),
    tibble(
      mae_assay_name = "amplicon", 
      se_object_name = "se_16S", 
      assay_label = "16S rRNA amplicon seq."
      ),
    tibble(
      mae_assay_name = "luminex", 
      se_object_name = "se_luminex", 
      assay_label = "Luminex"
      ),
    tibble(
      mae_assay_name = "flow", 
      se_object_name = "se_flow", 
      assay_label = "Flow cytometry"
      )
  ) |> 
  mutate(
    assay_label = assay_label |> fct_inorder()
  )

We will assemble the following tables (SE objects):

assay_list |> gt()

mae_assay_name	se_object_name	assay_label
mg	se_mg	Metagenomic
qPCR	se_qPRC	qPCR weekly swabs
qPCR_daily	se_daily_qPRC	qPCR home swabs
amplicon	se_16S	16S rRNA amplicon seq.
luminex	se_luminex	Luminex
flow	se_flow	Flow cytometry

mae_coldata_assays tables

available_samples <- 
  map(
    1:nrow(assay_list),
    function(i, assay_list){
      se <- get(assay_list$se_object_name[i])
      tibble(
        assay = assay_list$mae_assay_name[i],
        se_colname = se |> colnames(), 
        uid = se$uid, 
        sample_type = se$sample_type,
        control_type = se$control_type,
        )
    },
    assay_list = assay_list
  ) |> 
  bind_rows()

if (!all(available_samples$se_colname == available_samples$uid, na.rm = TRUE)) {
  stop("The uid column in the SE objects does not match their colnames")
}

mae_coldata_assays <-
  available_samples |>
  group_by(uid) |>
  summarize(
    n_sample_type = n_distinct(sample_type),
    n_control_type = n_distinct(control_type),
    sample_type = sample_type[1],
    control_type = control_type[1],
    assays = str_c(assay, collapse = ", ")
  )

if (any(mae_coldata_assays$n_sample_type > 1)) {
  warning("Some samples have multiple sample types")
}

if (any(mae_coldata_assays$n_control_type > 1)) {
  warning("Some samples have multiple control types")
}

mae_coldata_assays <- 
  mae_coldata_assays |> 
  select(uid, sample_type, control_type, assays) 

The mae_coldata_assays table contains the following columns:

mae_coldata_assays |> colnames()

[1] "uid"          "sample_type"  "control_type" "assays"      

mae_coldata_assays |> str()

tibble [4,185 × 4] (S3: tbl_df/tbl/data.frame)
 $ uid         : chr [1:4185] "068100004_0000" "068100004_1000" "068100004_1001" "068100004_1002" ...
 $ sample_type : chr [1:4185] "Clinical sample" "Clinical sample" "Clinical sample" "Clinical sample" ...
 $ control_type: chr [1:4185] "" "" "" "" ...
 $ assays      : chr [1:4185] "mg, qPCR, amplicon, luminex" "mg, qPCR, amplicon, luminex, flow" "qPCR_daily, amplicon" "qPCR_daily, amplicon" ...

mae_coldata_assays |> 
  dplyr::count(sample_type, control_type) |> 
  gt(caption = "Number of samples per sample type and control type") 

Number of samples per sample type and control type

sample_type	control_type	n
Clinical sample		3962
Clinical sample (other study)		7
Negative control	Nuclease-free water	46
Negative control	Unused swab + C2	47
Positive control	Mock 1	57
Positive control	Mock 2	57
Test sample		9

We also remove the columns sample_type and control_type from the colData of the individual assays to avoid conflicts when joining with the mae@colData table in downstream analyses.

for (i in 1:nrow(assay_list)) {
  se <- get(assay_list$se_object_name[i])
  if (("sample_type" %in% colnames(colData(se))) | ("control_type" %in% colnames(colData(se)))) {
    colData(se) <- colData(se)[, !colnames(colData(se)) %in% c("sample_type", "control_type")]
  }
  assign(assay_list$se_object_name[i], se)
}

MAE@colData: joining the visits and mae_coldata_assays tables

We first do a full join of the visits and mae_coldata_assays tables by uid. This will allow us to keep all visits, even those that do not have assay data, and to identify which visits are in the CRF data and which are in the assay data.

mae_coldata <- 
  # we first do a full join of the CRF visits and mae_coldata_assays by uids;
  dplyr::full_join(
    visits |> select(-starts_with("specimen_collection")) |> mutate(in_CRF = TRUE),
    mae_coldata_assays |> mutate(in_assays = TRUE) |> select(in_assays, everything()),
    by = join_by(uid)
  ) |> 
  mutate(
    in_CRF = in_CRF |> replace_na(FALSE),
    in_assays = in_assays |> replace_na(FALSE)
  )

# We notice that there are 5 "Clinical sample" entries for which we do have assay data but that did not appear in the CRF data:

mae_coldata |> 
  filter(is.na(visit_code), sample_type == "Clinical sample") 
  
# we fix these below

mae_coldata <- 
  mae_coldata |> 
  mutate(
    needs_fixing = is.na(pid) & (sample_type == "Clinical sample") & str_detect(uid, "[0-9]{9}_[0-9]{4}"),
    pid = 
      case_when(
        needs_fixing ~ str_remove(uid, "_[0-9]{4}"),
        TRUE ~ pid
      ),
    visit_code =
      case_when(
        needs_fixing ~ str_remove(uid, "[0-9]{9}_"),
        TRUE ~ visit_code
      ),
    study_day = 
      case_when(
        needs_fixing & (visit_code %in% c("0001", "0011")) ~ -7,
        needs_fixing & in_assays & (visit_code == "1000") ~ 0,
        TRUE ~ study_day
      ),
    visit_attended = 
      case_when(
        is.na(visit_attended) & needs_fixing ~ TRUE,
        TRUE ~ visit_attended
      ),
    visit_planned = 
      case_when(
        is.na(visit_planned) & needs_fixing ~ "Unplanned visit",
        TRUE ~ visit_planned
      ),
    visit_type = 
      case_when(
        is.na(visit_type) & needs_fixing ~ "Clinic",
        TRUE ~ visit_type
      )
  ) 

We then merge that table with the participants table to add the participant-invariant information (e.g., randomized, mITT, etc.) to the mae_coldata table.

# we add the participant-invariant information from the `participants` table
mae_coldata <- 
  mae_coldata |> 
  dplyr::left_join(
    participants,
    by = join_by(pid)
  )

mae_coldata <- mae_coldata |> select(-needs_fixing)

mae_coldata |> 
  dplyr::count(in_CRF, randomized, in_assays, sample_type, assays) |> 
  gt()

in_CRF	randomized	in_assays	sample_type	assays	n
FALSE	FALSE	TRUE	Clinical sample	mg, qPCR, amplicon, flow	2
FALSE	FALSE	TRUE	Clinical sample	qPCR_daily, amplicon	1
FALSE	NA	TRUE	Clinical sample (other study)	mg, qPCR	7
FALSE	NA	TRUE	Negative control	amplicon	70
FALSE	NA	TRUE	Negative control	mg, qPCR	12
FALSE	NA	TRUE	Negative control	qPCR	11
FALSE	NA	TRUE	Positive control	amplicon	91
FALSE	NA	TRUE	Positive control	mg	1
FALSE	NA	TRUE	Positive control	mg, qPCR	10
FALSE	NA	TRUE	Positive control	qPCR	12
FALSE	NA	TRUE	Test sample	qPCR	8
FALSE	NA	TRUE	Test sample	qPCR_daily	1
TRUE	FALSE	FALSE	NA	NA	414
TRUE	FALSE	TRUE	Clinical sample	mg, qPCR, amplicon	66
TRUE	FALSE	TRUE	Clinical sample	mg, qPCR, amplicon, luminex	5
TRUE	FALSE	TRUE	Clinical sample	qPCR, amplicon	1
TRUE	TRUE	FALSE	NA	NA	864
TRUE	TRUE	TRUE	Clinical sample	amplicon	4
TRUE	TRUE	TRUE	Clinical sample	flow	1
TRUE	TRUE	TRUE	Clinical sample	luminex, flow	1
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, amplicon	49
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, amplicon, flow	3
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, amplicon, luminex	491
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, amplicon, luminex, flow	346
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, luminex	2
TRUE	TRUE	TRUE	Clinical sample	mg, qPCR, qPCR_daily, amplicon, luminex, flow	2
TRUE	TRUE	TRUE	Clinical sample	qPCR, amplicon, luminex	1
TRUE	TRUE	TRUE	Clinical sample	qPCR, amplicon, luminex, flow	1
TRUE	TRUE	TRUE	Clinical sample	qPCR_daily, amplicon	2985
TRUE	TRUE	TRUE	Clinical sample	qPCR_daily, amplicon, luminex	1

Imputing study_day for daily visit_code that did not have CRF data

There were a few entries with daily visit_code that did not have a date in the CRF data. Consequently they also did not have study days (study_day) assigned.

mae_coldata |> 
  filter(
    is.na(study_day), 
    is.na(sample_type) | (sample_type == "Clinical sample")
    ) |> 
  dplyr::count(visit_type, assays, in_CRF, visit_attended, sample_type)

# A tibble: 5 × 6
  visit_type assays               in_CRF visit_attended sample_type         n
  <chr>      <chr>                <lgl>  <lgl>          <chr>           <int>
1 Clinic     <NA>                 TRUE   FALSE          <NA>               38
2 Home       qPCR_daily, amplicon TRUE   FALSE          Clinical sample   100
3 Home       qPCR_daily, amplicon TRUE   TRUE           Clinical sample     3
4 Home       <NA>                 TRUE   FALSE          <NA>              540
5 Home       <NA>                 TRUE   TRUE           <NA>                7

We impute the missing values using the study_days from “surrounding” visits.

# first, we create a "dictionary" with the most common study_days for each daily swab visit code
study_days_for_daily_codes <- 
  mae_coldata |> 
  dplyr::filter(visit_type == "Home") |> 
  dplyr::count(visit_code, study_day) |> 
  arrange(visit_code, -n) |> 
  group_by(visit_code) |> 
  slice_head(n = 1) |> 
  ungroup() |> 
  dplyr::rename(exp_study_day = study_day) |> 
  select(-n)

study_days_for_daily_codes |> 
  ggplot(aes(x = exp_study_day, y = visit_code)) +
  geom_point()

# In the plot above, we notice that visits 1501:1503 did not have "expected" study day, so we create it by expanding the time series
study_days_for_daily_codes <-
  study_days_for_daily_codes |> 
  mutate(
    exp_study_day = 
      case_when(
        is.na(exp_study_day) & (visit_code %in% c(1501:1503)) ~ 
          as.numeric(visit_code) - 1501 + 36,
        TRUE ~ exp_study_day
      )
  )

# visits from the participants with missing study_days
pid_with_missing_daily_study_days <- 
  # we first filter to retrieve the pids
  mae_coldata |> 
  filter(is.na(study_day)) |>  # , in_assays
  select(pid) |> distinct() |> 
  # we join with the coldata
  left_join(mae_coldata, by = join_by(pid)) |> 
  arrange(pid, visit_code) |> 
  # we join the study_day dictionary
  left_join(study_days_for_daily_codes) |> 
  mutate(diff = study_day - exp_study_day) |> # the typical difference for a participant between their study_day and the "expected" study_day (some participant started sampling at home one day early or late)
  group_by(pid) |> 
  fill(diff, .direction = "downup") |>  # impute when missing
  mutate(diff = ifelse(all(is.na(diff)), 0, diff)) |> # if still missing,we set to 0
  ungroup() |> 
  # we fix the study_day
  mutate(
    fixed_study_day = 
      case_when(
        is.na(study_day) & !is.na(exp_study_day) ~ exp_study_day + diff,
        TRUE ~ study_day
      )
  ) |> 
  select(uid, pid, visit_code, study_day, fixed_study_day, exp_study_day, diff, in_CRF, in_assays, crf_plates, assays)

# pid_with_missing_daily_study_days |> View()

mae_coldata <- 
  mae_coldata |> 
  left_join(
    pid_with_missing_daily_study_days |> select(uid, fixed_study_day), 
    by = join_by(uid)
    ) |> 
  mutate(
    study_day =
      case_when(
        is.na(study_day) ~ fixed_study_day,
        TRUE ~ study_day
      )
  ) |> 
  select(-fixed_study_day)

mae_coldata |> 
  filter(is.na(study_day), in_assays) |> 
  View()

# all is fixed

Creating the visit, visit_number, and study_week columns

visit_dict <- 
  bind_rows(
    tibble(
      visit = "Screening", 
      definition = "Screening visit for which we have the most assay data. If two screening visits have the same number of assays, we take the last one.", 
      possible_visit_codes = "0000, 0001, 0010, 0011",
      visit_number = "V0",
      study_week = -1
    ),
    tibble(
      visit = "Add. screening", 
      definition = 'If a participant has data for two screening visits, one is the "main" screening visit, and the other is labelled as an additional screening visit.', 
      possible_visit_codes = "0000, 0001, 0010, 0011",
      visit_number = "V0b",
      study_week = -2
      ),
    tibble(
      visit = "Day 0 (pre-MTZ)", 
      definition = "Visit before the metronidazole treatment (MTZ) starts.", 
      possible_visit_codes = "1000, 1011",
      visit_number = "V1",
      study_week = 0
      ),
    tibble(
      visit = "Week 1 (post-MTZ)", 
      definition = "Visit after the metronidazole treatment (MTZ) ends.", 
      possible_visit_codes = "1100",
      visit_number = "V2",
      study_week = 1
      ),
    tibble(
      visit = "Week 2 (post-SP)", 
      definition = "Visit after the study product (SP) ends.", 
      possible_visit_codes = "1200, 1211",
      visit_number = "V3",
      study_week = 2
      ),
    tibble(
      visit = "Week 3", 
      definition = "Visit one week after the study product (SP) ends.", 
      possible_visit_codes = "1300",
      visit_number = "V4",
      study_week = 3
      ),
    tibble(
      visit = "Week 4", 
      definition = "Visit two weeks after the study product (SP) ends.", 
      possible_visit_codes = "1400",
      visit_number = "V5",
      study_week = 4
      ),
    tibble(
      visit = "Week 5", 
      definition = "Visit three weeks after the study product (SP) ends.", 
      possible_visit_codes = "1500, 1511, 1512",
      visit_number = "V6",
      study_week = 5
      ),
    tibble(
      visit = "Week 7", 
      definition = "Visit five weeks after the study product (SP) ends.", 
      possible_visit_codes = "1700, 1711",
      visit_number = "V7",
      study_week = 7
      ),
    tibble(
      visit = "Week 9", 
      definition = "Visit seven weeks after the study product (SP) ends.", 
      possible_visit_codes = "1900, 1911",
      visit_number = "V8",
      study_week = 9
      ),
    tibble(
      visit = "Week 12", 
      definition = "Visit ten weeks after the study product (SP) ends.", 
      possible_visit_codes = "2120, 2121, 2122",
      visit_number = "V9",
      study_week = 12
    )
  ) |> 
  mutate(
    visit = visit |> fct_reorder(study_week)
  )

visit_dict |> 
  gt() |> 
  tab_header(
    title = "Visit codes and definitions"
  ) |> 
  cols_label(
    visit = "Visit",
    definition = "Definition",
    possible_visit_codes = "Possible visit codes",
    visit_number = "Visit number",
    study_week = "Study week"
  ) |> 
  cols_align(
    align = "left", columns = c(visit, definition, possible_visit_codes)
  )

Visit codes and definitions
Visit	Definition	Possible visit codes	Visit number	Study week
Screening	Screening visit for which we have the most assay data. If two screening visits have the same number of assays, we take the last one.	0000, 0001, 0010, 0011	V0	-1
Add. screening	If a participant has data for two screening visits, one is the "main" screening visit, and the other is labelled as an additional screening visit.	0000, 0001, 0010, 0011	V0b	-2
Day 0 (pre-MTZ)	Visit before the metronidazole treatment (MTZ) starts.	1000, 1011	V1	0
Week 1 (post-MTZ)	Visit after the metronidazole treatment (MTZ) ends.	1100	V2	1
Week 2 (post-SP)	Visit after the study product (SP) ends.	1200, 1211	V3	2
Week 3	Visit one week after the study product (SP) ends.	1300	V4	3
Week 4	Visit two weeks after the study product (SP) ends.	1400	V5	4
Week 5	Visit three weeks after the study product (SP) ends.	1500, 1511, 1512	V6	5
Week 7	Visit five weeks after the study product (SP) ends.	1700, 1711	V7	7
Week 9	Visit seven weeks after the study product (SP) ends.	1900, 1911	V8	9
Week 12	Visit ten weeks after the study product (SP) ends.	2120, 2121, 2122	V9	12

We first determine, for each participant, which is their main vs. their additional screening visit.

mae_coldata <- 
  mae_coldata |> 
  group_by(pid) |> 
  mutate(
    is_screening_visit = 
      case_when(
        visit_code %in% c("0000", "0001", "0010", "0011") ~ TRUE,
        TRUE ~ FALSE
      ),
    n_assays = ifelse(in_assays, str_split(assays, ", ") |> lengths(), 0),
    screening_visit_score = is_screening_visit * (n_assays + study_day/10)
  ) |> 
  arrange(-screening_visit_score) |> 
  mutate(
    screening_visit = 
      case_when(
        is_screening_visit & (row_number() == 1) ~ "main screening",
        is_screening_visit & (row_number() > 1) ~ "additional screening",
        TRUE ~ NA_character_
      )
  ) |> 
  ungroup() |> 
  select(-is_screening_visit, -n_assays, -screening_visit_score)

mae_coldata |> 
  dplyr::filter(screening_visit == "additional screening") |> 
  select(pid) |> distinct() |> 
  left_join(mae_coldata) |> 
  arrange(randomized |> desc(), pid, visit_code) |> 
  select(randomized, pid, visit_code, study_day, crf_plates, assays, screening_visit) |> 
  View()

mae_coldata |> 
  dplyr::filter(screening_visit == "additional screening") |> 
  select(pid) |> distinct() |> 
  left_join(mae_coldata) |> 
  arrange(randomized |> desc(), pid, visit_code) |> 
  select(randomized, pid, visit_code, study_day, crf_plates, assays, screening_visit) |> 
  dplyr::filter(randomized, (screening_visit == "main screening"), study_day == 0) |> 
  select(pid) |> distinct() |> 
  left_join(mae_coldata) |> 
  arrange(randomized |> desc(), pid, visit_code) |> 
  select(randomized, mITT, pid, visit_code, study_day, crf_plates, assays, screening_visit) |> 
  View()


# > check 068200020
# unclear - her main screening visit could be at day -3, but it might be at day 0, which would mean that her screening visit and pre-MTZ visit samples were collected on the same day
crf_raw$crf35 |> dplyr::filter(pid == "068200020") |> View()

mae_coldata |> 
  filter(!is.na(screening_visit)) |> 
  dplyr::count(screening_visit) 

# A tibble: 2 × 2
  screening_visit          n
  <chr>                <int>
1 additional screening    66
2 main screening         532

We next join the mae_coldata table with the visit_dict table to add the visit, visit_number, and study_week columns.

mae_coldata <- 
  mae_coldata |> 
  select(-any_of(c("PIPV", "visit", "visit_number", "study_week"))) |>
  left_join(
    visit_dict |> 
      mutate(visit_code = possible_visit_codes |> str_split(", ")) |>
      unnest(visit_code) |>
      mutate(
        screening_visit = 
          case_when(
            visit_number == "V0b" ~ "additional screening",
            visit_number == "V0" ~ "main screening",
            TRUE ~ NA_character_
            ),
        PIPV = visit_number != "V0b"
        ) |> 
      select(visit_code, screening_visit, PIPV, visit, visit_number, study_week)
  ) |> 
  mutate(PIPV = PIPV |> replace_na(FALSE)) |> 
  relocate(
    PIPV, visit, visit_number, study_week, .after = "visit_code"
  ) 

We check that study_days are unique for each visit_number at which assay samples were collected for each participant.

mae_coldata |> 
  group_by(pid, visit_number) |> 
  mutate(n_study_day = n_distinct(study_day[!is.na(assays)])) |>
  ungroup() |> 
  filter(PIPV, n_study_day > 1) |> 
  select(uid, pid, visit_code, study_day, assays) |> 
  group_by(pid) |> 
  gt(row_group_as_column = TRUE)

	uid	visit_code	study_day	assays
068200062	068200062_1500	1500	37	mg, qPCR, amplicon
	068200062_1511	1511	44	luminex, flow
068200087	068200087_1500	1500	40	mg, qPCR, amplicon, luminex
	068200087_1511	1511	43	flow

There are just two participants for which the dates don’t match and the different assays were collected at different dates. This may be important for integrative analyses later, so we leave as is for now.

Creating an assay to document specimen collection, and available data

• specimen_collection_swab (collected, not collected, unclear)
• specimen_collection_softcup (collected, not collected, unclear)
• specimen_collection_cytobrush (collected, not collected, unclear)
• mg
• qPCR
• amplicon
• luminex
• flow

sample_and_visit_info_assay <- 
  mae_coldata |> 
  select(uid, sample_type, in_CRF, in_assays, randomized) 

sample_and_visit_info_assay <- 
  sample_and_visit_info_assay |> 
  dplyr::left_join(
    visits |> select(uid, starts_with("specimen_collection")),
    by = join_by(uid)
  ) 

sample_and_visit_info_assay <-
  sample_and_visit_info_assay |>
  expand_grid(assay = assay_list$mae_assay_name) |>
  dplyr::left_join(
    available_samples |>
      select(uid, assay) |>
      mutate(value = TRUE),
    by = join_by(uid, assay)
  ) |>
  mutate(value = value |> replace_na(FALSE)) |>
  pivot_wider(names_from = assay, values_from = value)

# sample_and_visit_info_assay <- 
#   sample_and_visit_info_assay |> 
#   mutate(
#     mg_category = 
#       case_when(
#         !str_detect(sample_type, "Clinical sample") & mg ~ "Not a clinical sample",
#         is.na(randomized) & mg ~ "???",
#         !is.na(randomized) & !randomized & mg ~ "Additional sample from non-randomized participants",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & mg ~ "Expected data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & mg ~ "Unexpected data",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & !mg ~ "Missing data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & !mg ~ "Sample not collected",
#         TRUE ~ "ERROR"
#       ),
#     qPCR_category = 
#       case_when(
#         !str_detect(sample_type, "Clinical sample") & qPCR ~ "Not a clinical sample",
#         is.na(randomized) & qPCR ~ "???",
#         !is.na(randomized) & !randomized & qPCR ~ "Additional sample from non-randomized participants",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & qPCR ~ "Expected data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & qPCR ~ "Unexpected data",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & !qPCR ~ "Missing data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & !qPCR ~ "Sample not collected",
#         TRUE ~ "ERROR"
#       ),
#     amplicon_category = 
#       case_when(
#         !str_detect(sample_type, "Clinical sample") & amplicon ~ "Not a clinical sample",
#         is.na(randomized) & amplicon ~ "???",
#         !is.na(randomized) & !randomized & amplicon ~ "Additional sample from non-randomized participants",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & amplicon ~ "Expected data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & amplicon ~ "Unexpected data",
#         !is.na(specimen_collection_swab) & (specimen_collection_swab > 0) & !amplicon ~ "Missing data",
#         (is.na(specimen_collection_swab) | (specimen_collection_swab == 0)) & !amplicon ~ "Sample not collected",
#         TRUE ~ "ERROR"
#       ),
#     luminex_category = 
#       case_when(
#         !str_detect(sample_type, "Clinical sample") & luminex ~ "Not a clinical sample",
#         is.na(randomized) & luminex ~ "???",
#         !is.na(randomized) & !randomized & luminex ~ "Additional sample from non-randomized participants",
#         !is.na(specimen_collection_softcup) & (specimen_collection_softcup != "not collected") & luminex ~ "Expected data",
#         (is.na(specimen_collection_softcup) | (specimen_collection_softcup == "not collected")) & luminex ~ "Unexpected data",
#         !is.na(specimen_collection_softcup) & (specimen_collection_softcup == "collected") & !luminex ~ "Missing data",
#         (is.na(specimen_collection_softcup) | (specimen_collection_softcup == "not collected")) & !luminex ~ "Sample not collected",
#         TRUE ~ "ERROR"
#       )
#   )

# 
# The last 5 columns can take the following values:
# 
# - `expected` (swab collected and available data), 
# - `unexpected` (swab not collected but available data), 
# - `missing` (swab collected but no available data),
# - `maybe missing` (unclear if swab was collected and no available data),
# - `not collected` (swab not collected), 
# - `additional` (swab collected from a non-randomized participant), 
# - `NA` (not collected, no available data)
# 

se_sample_and_visit_info_assay <- 
  SummarizedExperiment(
    assays = list(
      sample_and_visit_info = 
        sample_and_visit_info_assay |> 
        select(-c(sample_type, in_CRF, in_assays, randomized)) |> 
        as.data.frame() |> 
        column_to_rownames("uid") |> 
        t()
    ),
    colData = 
      sample_and_visit_info_assay |> select(uid) |> 
      mutate(rownames = uid) |> as.data.frame() |> 
      column_to_rownames("rownames")
  )

Creating the MultiAssayExperiment object

SE_list <- 
  list(
    visit_and_sample_info = se_sample_and_visit_info_assay,
    mg = se_mg,
    qPCR = se_qPRC,
    qPCR_daily = se_daily_qPRC,
    amplicon = se_16S,
    luminex = se_luminex,
    flow = se_flow
  )


mae <- 
  MultiAssayExperiment::MultiAssayExperiment(
    experiments = MultiAssayExperiment::ExperimentList(SE_list),
    colData = 
      mae_coldata |> select(-in_CRF, -in_assays) |> 
      as.data.frame() |> mutate(rownames = uid) |> 
      column_to_rownames("rownames"),
    metadata = list(
      study = "VIBRANT",
      data_source = "actual study data (not simulated)",
      data_status = "partially QCed",
      date = today(),
      colData_dictionary = participants_variable_dictionary,
      crf_plates_description = crf_plates_dictionary,
      crf_data_raw = crf_raw,
      crf_data_clean = crf_clean, 
      participant_crfs_merged = participant_crfs_merged,
      visits_crfs_merged = visits_crfs_merged,
      exposures = exposures
    )
  )

mae

A MultiAssayExperiment object of 7 listed
 experiments with user-defined names and respective classes.
 Containing an ExperimentList class object of length 7:
 [1] visit_and_sample_info: SummarizedExperiment with 9 rows and 5463 columns
 [2] mg: SummarizedExperiment with 779 rows and 996 columns
 [3] qPCR: SummarizedExperiment with 16 rows and 1029 columns
 [4] qPCR_daily: SummarizedExperiment with 16 rows and 2990 columns
 [5] amplicon: SummarizedExperiment with 817 rows and 4119 columns
 [6] luminex: SummarizedExperiment with 48 rows and 850 columns
 [7] flow: SummarizedExperiment with 11 rows and 356 columns
Functionality:
 experiments() - obtain the ExperimentList instance
 colData() - the primary/phenotype DataFrame
 sampleMap() - the sample coordination DataFrame
 `$`, `[`, `[[` - extract colData columns, subset, or experiment
 *Format() - convert into a long or wide DataFrame
 assays() - convert ExperimentList to a SimpleList of matrices
 exportClass() - save data to flat files

Available assays

mae_sample_map <- 
  mae@sampleMap |> as_tibble() |> dplyr::filter(assay %in% assay_list$mae_assay_name) |>
  select(uid = primary, assay) |> 
  arrange(uid) |> 
  dplyr::left_join(assay_list |> dplyr::rename(assay = mae_assay_name), by = join_by(assay)) |> 
  dplyr::right_join(
    mae@colData |> as.data.frame() |> as_tibble() |> 
      select(uid, pid, visit_code, PIPV, visit, visit_type, sample_type, location, randomized, arm), 
    by = join_by(uid)
    )

All samples

mae_sample_map |> 
  dplyr::filter(!is.na(assay)) |> 
  group_by(uid) |> mutate(n = n()) |> ungroup() |> 
  mutate(uid = uid |> fct_reorder(-n)) |>
  ggplot() +
  aes(x = uid, y = assay_label |> fct_rev(), fill = assay_label) +
  geom_tile() +
  facet_grid(. ~ sample_type, scales = "free_x", space = "free_x") +
  scale_x_discrete("Sample IDs", breaks = NULL) +
  ylab("") +
  scale_fill_brewer("Assay", type = "qual", palette = 7) + # 3
  theme(
    axis.text.x = element_text(angle = 45, hjust = 1),
    strip.text.x = element_text(angle = 90),
    legend.position = "bottom"
    )

Longitudinal availability for all visits

mae_sample_map |> 
  dplyr::filter(!is.na(assay)) |> 
  # filter(visit_code %in% c("0000", seq(1000, 1900, by = 100), "2120")) |> 
  ggplot() +
  aes(x = assay_label, y = pid, fill = assay_label) +
  geom_tile() +
  facet_grid(
    ifelse(randomized, "Randomized", "Not randomized") + location ~ visit_code, 
    scales = "free_y", space = "free_y"
    ) + # sample_category
  scale_y_discrete("Participants IDs") + #, breaks = "NULL")  +
  xlab("") +
  scale_fill_brewer("Assay", type = "qual", palette = 7) + # 3
  theme(
    axis.text.x = element_blank(), #element_text(angle = 45, hjust = 1),
    strip.text.y = element_text(angle = 0),
    strip.text.x = element_text(angle = 90),
    legend.position = "bottom"
    )

For the weekly visits of all participants

mae_sample_map |> 
  dplyr::filter(!is.na(assay)) |> 
  dplyr::filter(visit_type == "Clinic") |> 
  ggplot() +
  aes(x = assay_label, y = pid, fill = assay_label) +
  geom_tile() +
  facet_grid(
    location + ifelse(randomized, "Randomized", "Not randomized")  + arm  ~ visit_code, 
    scales = "free_y", space = "free_y"
    ) + # sample_category
  scale_y_discrete("Participants IDs") + #, breaks = "NULL")  +
  xlab("") +
  scale_fill_brewer("Assay", type = "qual", palette = 7) + # 3
  theme(
    axis.text.x = element_blank(), #element_text(angle = 45, hjust = 1),
    strip.text.y = element_text(angle = 0),
    strip.text.x = element_text(angle = 90),
    legend.position = "bottom"
    )

For the weekly visits of randomized participants

mae_sample_map |> 
  dplyr::filter(randomized) |> 
  dplyr::filter(!is.na(assay)) |> 
  dplyr::filter(visit_type == "Clinic") |> 
  ggplot() +
  aes(x = assay_label, y = pid, fill = assay_label) +
  geom_tile() +
  facet_grid(
    location + ifelse(randomized, "Randomized", "Not randomized")  + arm  ~ visit_code, 
    scales = "free_y", space = "free_y"
    ) + # sample_category
  scale_y_discrete("Participants IDs") + #, breaks = "NULL")  +
  xlab("") +
  scale_fill_brewer("Assay", type = "qual", palette = 7) + # 3
  theme(
    axis.text.x = element_blank(), #element_text(angle = 45, hjust = 1),
    strip.text.y = element_text(angle = 0),
    strip.text.x = element_text(angle = 90),
    legend.position = "bottom"
    )

mae_sample_map |> 
  dplyr::filter(randomized) |> 
  dplyr::filter(!is.na(assay)) |> 
  dplyr::filter(PIPV) |> 
  ggplot() +
  aes(x = assay_label, y = pid, fill = assay_label) +
  geom_tile() +
  facet_grid(
    location + arm  ~ visit, 
    scales = "free_y", space = "free_y"
    ) + # sample_category
  scale_y_discrete("Participants IDs") + #, breaks = "NULL")  +
  xlab("") +
  scale_fill_brewer("Assay", type = "qual", palette = 7) + # 3
  theme(
    axis.text.x = element_blank(), #element_text(angle = 45, hjust = 1),
    strip.text.y = element_text(angle = 0),
    strip.text.x = element_text(angle = 90),
    legend.position = "bottom"
    )

Saving the MultiAssayExperiment objects

We export the full MultiAssayExperiment object as an .rds file for further analyses.

saveRDS(
  mae, 
  str_c(
    get_02_output_dir(),
    "mae_full_", today() |> str_remove_all("-"), ".rds"
  )
)